November 19, 2026 · prices at the close · estimates collected before it
Open source and software valuations
Snowflake is the clearest case here of a demanding price meeting a substitute that already takes paid work; Datadog is second. Snowflake’s price needs 29% sales growth a year through January 2034, and New Relic says it cut data-platform spending 35–52% by moving off it. Cloudflare and JFrog need the most growth, 50% and 37% a year against forecasts of 30% and 19%, but open source replaces little of what they sell. Elastic, GitLab and MongoDB have the most mature substitutes and are priced for less growth than analysts forecast. AI has not moved existing bills, but agents now choose the software for much new work and they choose open defaults, which threatens the growth these prices need. None of it shows in reported results: all seven beat earnings estimates in each of the last four quarters.
Price against exposure
analysts’ forecast for the second yeargrowth the price needs in years three to seventhe same at a 35% margin
| Company | Analysts’ growth next two years | Growth the price needs¹ | Sales a substitute can reach² | Share price, middle case³ | Paid work seen leaving | Cash margin today⁴ | EV ÷ sales | |
|---|---|---|---|---|---|---|---|---|
| Price needs far more growth than analysts forecast | ||||||||
| CloudflareNET · $303.69 | 32% / 30% | 50%42% at a 35% margin | 10% | −4%−11% severe | None | −9% | 36.5× | |
| JFrogFROG · $109.27 | 20% / 19% | 37%29% at a 35% margin | 25% | −10%−26% severe | None; a $76M contract arrived | 7% | 18.7× | |
| Price needs today’s growth to last seven more years | ||||||||
| DatadogDDOG · $166.46 | 22% / 23% | 26%19% at a 35% margin | 30% | −12%−32% severe | One small case | 6% | 12.3× | |
| SnowflakeSNOW · $268.56 | 29% / 29% | 29%21% at a 35% margin | 40% | −15%−40% severe | One large case, no dollars given | −8% | 14.7× | |
| Price needs less growth than analysts forecast | ||||||||
| GitLabGTLB · $43.68 | 16% / 16% | 12%6% at a 35% margin | 35% | −14%−39% severe | One case, not clearly a paid one | −2% | 5.3× | |
| MongoDBMDB · $297.01 | 20% / 20% | 16%9% at a 35% margin | 25% | −9%−25% severe | One case, from 2018 | 3% | 7.1× | |
| ElasticESTC · $76.11 | 15% / 15% | 4%0% or less at a 35% margin | 40% | −16%−42% severe | None; one customer arrived | 3% | 3.5× | |
¹ Sales growth in forecast years three to seven that makes discounted cash flow equal the share price, taking analysts’ sales and free cash flow for years one and two, a 25% cash margin by year ten, growth fading to 3%, and a 10% discount rate. ² My judgment of the share of sales in products with a working open-source substitute. ³ Change in share price if 15% of that work leaves by August 2029, the rest is discounted 5% and 5% is won back in other products. Severe: 35% leaves, 10% discount, 2% won back. Lost volume costs its gross margin; costs are otherwise held fixed. ⁴ Free cash flow less stock pay, last twelve months, as a share of sales.
What the prices need
After stock pay the seven keep between −9% and 7% of sales as cash today. The model gives every one of them 25% by year ten; Adobe kept about 33% in fiscal 2025. At 35% each hurdle falls about 7 points. The discount rate is 10% against a ten-year Treasury yield of 5.04%.
Cloudflare last grew 50% in 2021, on $656M of sales; its price needs that pace starting from $5.10B. JFrog last grew 37% in 2021, on $207M of sales; it needs it from $973M. Snowflake and Datadog grew 29% and 28% last fiscal year and need about that through January 2034 and December 2033. Elastic’s price holds if growth falls to 4%.
What open source can take
A substitute that works is not yet a smaller bill. The table lists what open source replaces at each company, what it does not, and the best named case, each checked against its source. Only Aithon itemizes a bill that left one of these vendors, and it is $120,000 a year. New Relic is the one large case and gives percentages only. Lumen, Loadsmart and JFrog’s telecom customer moved toward the incumbent.
| Company | What open source replaces | What it does not | Named case, and who published it |
|---|---|---|---|
| Snowflake | Warehouse compute, the bulk of sales (storage was about 11% when last disclosed). Jobs and dashboards run on another engine over the same Iceberg tables. | Data sharing, governance, burst capacity. | New Relic, April 2026: 1,000+ datasets to an Iceberg stack, annual spend down 35–52%. Its own engineering blog; no dollars, and it does not say Snowflake is switched off. |
| Databricks | The SQL warehouse, about 21% of sales, plus scheduled transformations and the managed Postgres mirror. | Unity Catalog, managed Spark, AI services. | Aithon: 165 GB to Postgres and dbt, $10,000 a month to an estimated $700–1,300, on a database it already paid for. Its consultant. Other direction: Lumen, 133 TB from on-premises Cloudera onto Databricks in nine months, compute down 30–40%. Databricks. |
| Datadog | Log storage and retention, through OpenTelemetry into ClickHouse or Loki. Logs were about a quarter of sales when last disclosed. | Tracing, application monitoring, the incident workflow. | Trunk: self-run ClickHouse is 15–20× cheaper by one engineer’s estimate, covering all traffic where Datadog covered 15%. ClickHouse. |
| JFrog | The container registry (Harbor). Sales below the Enterprise+ tier are 39%. | Other package formats, provenance, release policy. | None leaving. A telecom that first came over from Sonatype Nexus signed a $76M multiyear cloud contract. JFrog. |
| MongoDB | The next document application, started on Postgres or DocumentDB. | Running Atlas applications: indexes, failover, search. Atlas is 73% of sales. | The Guardian, 2018: 2.3M items from self-run MongoDB, with a paid support contract, to Postgres. Its own blog. |
| GitLab | Git hosting and simple pipelines (Forgejo). Tiers below Ultimate are 41% of recurring revenue. | Approvals, security scanning, delivery pipelines. | Smartclip: 150 repositories in three months, for security and simplicity. It does not say it paid GitLab. Its own blog. |
| Elastic | Search and log clusters (OpenSearch, which AWS sells managed). Self-managed subscriptions are 43% of sales, Elastic Cloud 51%. | Relevance tuning, security content. | None leaving Elastic. Halodoc moved from AWS’s Elasticsearch service to AWS OpenSearch, cost down 36%. Loadsmart moved from AWS to Elastic Cloud; median latency went from 90 ms to 40 ms. Their own blogs. |
| Cloudflare | The Workers runtime, which Cloudflare publishes as workerd. | The network in 340+ cities, attack protection, delivery. | None. |
| Anthropic | Routine inference, sent to open-weight models. | Hard coding work, reliability, latency. | OpenRouter: DeepSeek’s share of tokens on its network rose from 9% to 18% in the first half of 2026; its share of spending rose less. |
Scale. A thousand exits the size of Aithon’s would take $120M a year from Databricks. Its middle case needs $1.01B. Databricks has over 1,000 customers paying more than $1M a year: one of those renewals, with the whole bill before and after, would settle more than every case above.
Precedent. Elastic has faced the strongest version since 2021, when AWS forked its code as OpenSearch and sold it as a service. Since then Elastic’s growth has fallen from 42% to 17%; without a fork, MongoDB’s fell from 48% to 23% and Datadog’s from 63% to 28%. Elastic’s sales rose every year. It is now the cheapest of the seven, at 3.5 times sales.
What AI changes
Open substitutes are a decade old. What changed is the cost of code, and so far it has moved new work, not existing bills.
Rewriting is nearly free. Bun ported 535,496 lines from Zig to Rust in 11 days for about $165,000 of model usage, against its own estimate of three engineers for a year. Asana moved its tests off an abandoned framework for $12,000. One Cloudflare engineer rebuilt 94% of the Next.js API in a week for $1,100. Anthropic replaced git with a version-control system that seven people built in five months. The vendors give converters away to poach each other: Databricks’ free Lakebridge takes Snowflake as a source and claims to automate up to 80% of a migration.
Proving the result is not. Databricks puts validation at 50–60% of migration effort. A buyer moving $1M a year of work to a replacement that costs $300k in cloud bills and $350k in people saves $350k a year, so a $500k migration pays back in 17 months. Halve the migration and it pays back in 9. I found no named case of agents moving production work off any of the eight companies.
Agents choose the software for new work, and they choose open defaults. Agents create 70% of new Supabase databases and start more than 40% of GitHub pull requests, up from 17% in March. In a study of Claude Code’s choices, Postgres took 53% of database picks and MongoDB none, though MongoDB was named in a third of answers; GitHub Actions took every build-pipeline pick; and code written on the spot was the most common choice overall, at 21%. In another test, three models setting up logging installed Grafana’s stack in 12 of 30 answers, OpenTelemetry in 10 and Datadog in none. These prices need growth, growth comes from new work, and agents decide more of it each quarter.
AI also buys from these companies. AI products supplied about half of Snowflake’s recent acceleration. Datadog’s AI-native customers, its largest among them, added high-single-digit points to its 35% growth; that customer cut usage last quarter, and guidance assumes it falls to its contract minimum. Databricks had a $1.4B AI sales pace in February. GitLab’s usage-priced products run at $40M a year, about 4% of sales; the rest is mostly sold by the seat.
Databricks
Private, valued at $190B in August on a sales pace above $7B growing more than 80%; the warehouse product alone is above $1.5B and growing more than 100%. The reference path takes growth down to 50%, 35% and 25%, reaching $10.50B in August 2027 and $17.72B in August 2029, with the warehouse at 30% of sales by then. Each row is the change from that path.
| Future | What is assumed for August 2029 | Sales, Aug 2027 | Sales, Aug 2029 | Operating profit, Aug 2029 | Sales pace reached |
|---|---|---|---|---|---|
| Open access wins | 8% of warehouse work and 1% of other work leaves; prices fall 3% and 0%; new sales add 5% | +$15M | +$190M+1.1% | +$22M to +$106M | $17.91B |
| Two engines share the data | 15% of warehouse work and 2% of other work leaves; prices fall 5% and 0%; new sales add 1.5% | −$218M | −$1.01B−5.7% | −$811M to −$616M | $16.71B |
| Buyers negotiate lower prices | 5% of warehouse work and 1% of other work leaves; prices fall 15% and 3%; new sales add 1% | −$324M | −$1.34B−7.6% | −$1.29B to −$1.23B | $16.38B |
| Pipelines leave too | 35% of warehouse work and 8% of other work leaves; prices fall 10% and 2%; new sales add 1% | −$500M | −$3.25B−18.3% | −$2.58B to −$1.91B | $14.47B |
I weight the second row most: it has an itemized customer bill, a supported way to run two engines on one catalog, and a reason to keep the rest of the platform. In all four futures sales more than double, to $14.47B–$17.91B, so open source moves the valuation less than margin does. At 30 times earnings, $190B needs $6.33B of profit, 38% of middle-case sales; a 10% annual return to August 2029 needs $8.43B, or 50%.
Anthropic
Listed November 19 and worth about $1.25T that afternoon, 35 times trailing revenue. Quarterly revenue rose from $1.4B to $17B in a year. The prospectus discloses $900.9B of signed compute and lease commitments, $70.7B of it due within twelve months, so a price cut on routine inference lands on fixed costs. No options were listed on the first day.
What options price
Options price a move of 15–22% in each stock by December 18, against 8.5% for the software ETF. The whole middle case is 4–16% of each stock and arrives through 2029. A put bought today therefore needs most of a three-year effect in 29 days; for MongoDB and Cloudflare the entire effect falls short of break-even.
| Stock | Move priced by Dec 18 | Dec 18 put, strike and ask | Fall needed to break even | Middle case, whole effect | Share of it needed in 29 days |
|---|---|---|---|---|---|
| SNOW | 18.9% | $270.00 at $26.30 | 9.3% | 15.4% | 60% |
| DDOG | 15.5% | $165.00 at $12.35 | 8.3% | 11.8% | 70% |
| ESTC | 20.9% | $75.00 at $7.70 | 11.6% | 15.8% | 73% |
| GTLB | 17.6% | $42.50 at $3.50 | 10.7% | 14.2% | 75% |
| FROG | 16.7% | $110.00 at $9.90 | 8.4% | 9.8% | 86% |
| MDB | 22.4% | $300.00 at $35.85 | 11.1% | 9.4% | 118% |
| NET | 16.3% | $300.00 at $23.15 | 8.8% | 4.1% | 218% |
MongoDB and GitLab report December 1 and Snowflake December 2, inside the December contracts: the priced move rises from 8.7% to 19.8% for MongoDB, 7.3% to 16.4% for Snowflake and 7.5% to 15.9% for GitLab between the November 27 and December 4 expiries. Elastic reported after the close on November 19 ($76.11 at the close, $72.06 after hours); its quotes predate the report. The last listed expiry in every chain is January 19, 2029, seven months before the scenarios end.
Earnings
Analysts’ estimates for the next full fiscal year. Net income and EPS include stock pay; adjusted EPS excludes it. JFrog is the odd one: analysts forecast a $13M loss, though each new dollar of sales brought 33¢ of operating profit over the past year, which would give $32M of profit.
| Company, fiscal year | Sales | Growth | Net income | EPS | Adjusted EPS | Price ÷ adjusted EPS |
|---|---|---|---|---|---|---|
| CloudflareDec 2027 | $3.92B | 32.2% | $68M | $0.17 | $1.86 | 163× |
| JFrogDec 2027 | $819M | 20.2% | −$13M | −$0.12 | $1.33 | 82× |
| DatadogDec 2027 | $5.53B | 22.3% | $386M | $0.95 | $3.18 | 52× |
| SnowflakeJan 2028 | $8.17B | 29.3% | −$511M | −$1.57 | $3.08 | 87× |
| GitLabJan 2028 | $1.31B | 15.7% | −$61M | −$0.35 | $1.04 | 42× |
| MongoDBJan 2028 | $3.62B | 19.5% | $117M | $1.39 | $7.92 | 38× |
| ElasticApr 2028 | $2.30B | 14.8% | $89M | $0.80 | $3.98 | 19× |
Cloudflare Dec 2025 to Dec 2028, cash after stock pay, latest quarter
| $ | Dec 2025actual | Dec 2026estimate | Dec 2027estimate | Dec 2028estimate |
|---|---|---|---|---|
| Sales | $2.17B | $2.96B | $3.92B | $5.10B |
| Net income | −$102M | −$191M | $68M | $221M |
| EPS | −$0.29 | −$0.53 | $0.17 | $0.54 |
| Adjusted EPS | $0.93 | $1.35 | $1.86 | — |
| Free cash flow | $261M | $349M | $585M | $888M |
| Stock pay | $451M | $615M | $818M | $1.07B |
Twelve months to September 2026: sales $2.73B, gross margin 72%, free cash flow $322M, stock pay $565M, leaving −$243M (−9% of sales). September quarter: revenue up 39%; gross margin 71.2%, from 74.0%; free cash flow margin 10%, from 13%. Nine months of restructuring and other charges came to $162M.
JFrog Dec 2025 to Dec 2028, cash after stock pay, latest quarter
| $ | Dec 2025actual | Dec 2026estimate | Dec 2027estimate | Dec 2028estimate |
|---|---|---|---|---|
| Sales | $532M | $682M | $819M | $973M |
| Net income | −$72M | −$13M | −$13M | $22M |
| EPS | −$0.62 | −$0.09 | −$0.12 | $0.02 |
| Adjusted EPS | $0.82 | $1.15 | $1.33 | — |
| Free cash flow | $142M | $198M | $235M | $297M |
| Stock pay | $157M | $169M | $193M | $221M |
Twelve months to September 2026: sales $642M, gross margin 78%, free cash flow $207M, stock pay $162M, leaving $45M (7% of sales). September quarter: sales of $179M, up 31% and 9% above the guidance midpoint; cloud revenue up 60%; net retention 124%; 112 customers above $1M a year, from 71. Enterprise+ was 61% of revenue. The provider’s FY2028 figures disagree with each other ($22M of net income, $0.02 of EPS, on 125M shares).
Datadog Dec 2025 to Dec 2028, cash after stock pay, latest quarter
| $ | Dec 2025actual | Dec 2026estimate | Dec 2027estimate | Dec 2028estimate |
|---|---|---|---|---|
| Sales | $3.43B | $4.52B | $5.53B | $6.79B |
| Net income | $108M | $280M | $386M | $587M |
| EPS | $0.31 | $0.75 | $0.95 | $1.38 |
| Adjusted EPS | $2.05 | $2.74 | $3.18 | — |
| Free cash flow | $915M | $1.19B | $1.52B | $1.97B |
| Stock pay | $751M | $905M | $1.09B | $1.32B |
Twelve months to September 2026: sales $4.28B, gross margin 80%, free cash flow $1.11B, stock pay $860M, leaving $250M (6% of sales). September quarter: sales of $1.197B, up 35% and 5% above the guidance midpoint; backlog up 38%; net retention in the low 120s. Growth outside AI-native customers reached the high 20s. The largest customer cut usage, and fourth-quarter guidance (25–26% growth) assumes it falls to its contractual minimum.
Snowflake Jan 2026 to Jan 2029, cash after stock pay, latest quarter
| $ | Jan 2026actual | Jan 2027estimate | Jan 2028estimate | Jan 2029estimate |
|---|---|---|---|---|
| Sales | $4.68B | $6.32B | $8.17B | $10.54B |
| Net income | −$1.33B | −$894M | −$511M | −$24M |
| EPS | −$3.95 | −$2.50 | −$1.57 | −$0.26 |
| Adjusted EPS | $1.25 | $2.23 | $3.08 | — |
| Free cash flow | $1.12B | $1.43B | $1.97B | $2.88B |
| Stock pay | $1.60B | $1.78B | $2.07B | $2.44B |
Twelve months to July 2026: sales $5.43B, gross margin 67%, free cash flow $1.20B, stock pay $1.64B, leaving −$447M (−8% of sales). July quarter: product revenue up 37%, net revenue retention 126%, contracted backlog $9.0B, up 30%. The full-year product guide rose from $5.84B to $6.07B. Reports December 2.
GitLab Jan 2026 to Jan 2029, cash after stock pay, latest quarter
| $ | Jan 2026actual | Jan 2027estimate | Jan 2028estimate | Jan 2029estimate |
|---|---|---|---|---|
| Sales | $955M | $1.13B | $1.31B | $1.52B |
| Net income | −$56M | −$89M | −$61M | −$41M |
| EPS | −$0.34 | −$0.53 | −$0.35 | −$0.25 |
| Adjusted EPS | $0.96 | $0.87 | $1.04 | — |
| Free cash flow | $222M | $222M | $281M | $374M |
| Stock pay | $215M | $241M | $265M | $295M |
Twelve months to July 2026: sales $1.06B, gross margin 86%, free cash flow $214M, stock pay $230M, leaving −$16M (−2% of sales). July quarter: net retention 117%, twelve-month backlog up 20%. A restructuring ($30–35M of charges) and new Flex contracts change how customers commit to seats and usage. Reports December 1.
MongoDB Jan 2026 to Jan 2029, cash after stock pay, latest quarter
| $ | Jan 2026actual | Jan 2027estimate | Jan 2028estimate | Jan 2029estimate |
|---|---|---|---|---|
| Sales | $2.46B | $3.03B | $3.62B | $4.35B |
| Net income | −$71M | $60M | $117M | $225M |
| EPS | −$0.88 | $0.65 | $1.39 | $2.61 |
| Adjusted EPS | $4.97 | $6.55 | $7.92 | — |
| Free cash flow | $493M | $614M | $768M | $1.02B |
| Stock pay | $550M | $585M | $635M | $697M |
Twelve months to July 2026: sales $2.78B, gross margin 73%, free cash flow $652M, stock pay $564M, leaving $88M (3% of sales). July quarter: Atlas was 73% of revenue and grew 29%. Targets from the September investor day: at least 20% growth, mid-20s for Atlas, and profit from fiscal 2027. Reports December 1.
Elastic Apr 2026 to Apr 2029, cash after stock pay, latest quarter
| $ | Apr 2026actual | Apr 2027estimate | Apr 2028estimate | Apr 2029estimate |
|---|---|---|---|---|
| Sales | $1.74B | $2.00B | $2.30B | $2.64B |
| Net income | $368M | $30M | $89M | $161M |
| EPS | $3.43 | $0.26 | $0.80 | $1.36 |
| Adjusted EPS | $2.57 | $3.36 | $3.98 | — |
| Free cash flow | $322M | $424M | $532M | $605M |
| Stock pay | $298M | $328M | $363M | $403M |
Twelve months to July 2026: sales $1.80B, gross margin 76%, free cash flow $349M, stock pay $303M, leaving $46M (3% of sales). October quarter, reported after these estimates were collected: revenue of $496M, up 17%. Last fiscal year’s $368M of net income included a deferred-tax benefit of about $435M.
Igloo’s benchmark
The Snowflake and Databricks cases lean on one engine’s own numbers. Igloo packages DuckDB, Iceberg tables and the PostgreSQL wire protocol as a self-hosted warehouse; its releases so far are release candidates. Its published benchmark runs the 22 TPC-H queries on 100 GB one at a time, with warm data caches, result caches off, at list prices. Across ten pairings it reports medians of 3 times the speed at one twenty-fifth of the compute cost. The medium row below sets a 64-core Igloo machine against each medium warehouse; on 16 cores Igloo is 1.1 times Snowflake’s speed and 0.8 times Databricks’. Cost counts query seconds only, so idle time, concurrency, staffing and negotiated rates are outside it. I did not rerun it. PostHog moved its own warehouse product off DuckDB at 2 PB after five months.
| Medium warehouse, 22 queries | Time | Compute cost |
|---|---|---|
| Igloo | 15.4 s | $0.010 |
| Databricks SQL Serverless | 48.1 s | $0.225 |
| Snowflake Gen2 | 61.7 s | $0.278 |
Method and limits
Price hurdle. Analysts’ sales and free cash flow for the rest of this fiscal year and the next two (MarketScreener; Yahoo Finance for adjusted EPS), less projected stock pay and after-tax interest income. From year three, sales grow at the solved rate through year seven, then fade to 3% by year ten, while the cash margin moves in equal steps from its year-two level to 25%. Cash is discounted at 10% with 3% terminal growth, net cash is added, and the share count is held fixed because stock pay is already charged as a cost. A second, separately written calculation reproduces every price from its solved rate.
Scenario prices. The exposed share times a scenario’s net loss is a lasting change in sales that reaches a third of its size by August 2027 and all of it by August 2029. Lost volume costs the trailing gross margin, a price cut costs a dollar per dollar, tax is 21%, and no other cost adjusts, which overstates the damage to a company that cuts spending in response.
Limits. The exposed shares and scenario rates are my judgments, and one set of four scenarios is applied to every company. Forecasts were collected on November 19 before the close; Elastic’s predate its report that evening. Stock-pay forecasts extend each company’s last two years. The cases come from engineering blogs and vendor marketing. Databricks’ product mix and margins are private, and its starting figures are the floors it disclosed.
What would change the conclusion. A customer paying Snowflake more than $1M a year publishing its whole bill before and after a renewal; a named Enterprise+ customer leaving Artifactory; Datadog’s net retention falling while log volume grows. Every quote and assumption downloads from the options model.